**Course:** Hermes Agent — From Zero to Autonomous Agent
**Module:** 1 of 8
**Reading time:** ~15 minutes
**Difficulty:** Beginner — no prior AI agent experience needed
TL;DR
| **What it is** | An **open-source AI agent framework by Nous Research** that runs in your terminal, connects to 20+ LLM providers, and gets smarter over time through a built-in learning loop. |
|---|---|
| **Not a chatbot** | Unlike ChatGPT or Claude, Hermes has **persistent memory**, a **skills system** that self-improves, **tool calling** (file system, terminal, web, APIs), and can run **autonomously on a schedule** from a $5 VPS. |
| **Key superpowers** | Provider-agnostic, self-improving skills, persistent memory, multi-platform (Telegram, Discord, Slack, Signal + 15+ more), cron jobs, subagent delegation, MCP server support. |
| **Who it's for** | Developers, sysadmins, power users, researchers, home automators — anyone who wants an AI that does work, not just answers questions. |
| **License** | MIT — free forever, self-hosted, no data leaves your infra unless you choose. |
The Problem: Chatbots Answer Questions. Hermes Gets Things Done.
Let's be honest — you've used ChatGPT, Claude, or Gemini. You've been impressed. You've asked it to write code, explain concepts, draft emails. And then… you copy-paste the output into some other tool, run the command yourself, save the file manually, switch to Slack to share it.
That's the **chatbot pattern**: LLM in a browser tab, disconnected from your actual workflow.
Hermes Agent breaks that pattern completely. Instead of a web chat window, it lives inside your **terminal** — connected to your file system, your shell, your browser, your APIs, your messaging apps. It doesn't just *tell you* the command to run — it **runs it**. It doesn't just *write* a script — it **executes it**, **tests it**, and **fixes it** if it fails.
And then it remembers what it learned for next time.
What Hermes Agent Actually Is
Hermes Agent is an **open-source, autonomous AI agent framework** built by [Nous Research](https://nousresearch.com) — the same lab that trained the Hermes, Nomos, and Psyche language models. It's not a product you subscribe to; it's software you **self-host** on your own machine (or a $5 VPS, or a GPU cluster, or serverless infrastructure).
Here's what makes it fundamentally different from a web chatbot:
1. It Runs in Your Terminal
| ChatGPT / Claude | Hermes Agent | |
|---|---|---|
| **Interface** | Web browser / mobile app | Terminal CLI, TUI, or messaging apps |
| **Execution** | Tells you what to do | Does it for you |
| **Your data** | On their servers | On your machine |
| **Tools** | Browsing (limited) | 70+ built-in tools: shell, files, web, browser, git, APIs |
| **Memory** | Session-only (or thin persistent) | Persistent, curated, cross-session |
| **Autonomy** | Responds when prompted | Runs on cron, spawns subagents, acts proactively |
When you launch Hermes, you get a `$` prompt in your terminal — just like a shell. But instead of running commands directly, you're talking to an AI that **has access** to your shell, file system, and a growing toolkit. Type a request, and Hermes:
1. **Thinks** about what needs to happen
2. **Calls tools** — writes files, runs terminal commands, searches the web, reads your codebase
3. **Iterates** — if the command fails, it reads the error and fixes it
4. **Delivers** the result
2. Provider-Agnostic: No Lock-In
This is one of the most liberating features. Hermes works with **20+ LLM providers** out of the box, and you can switch between them with a single command:
hermes model # Interactive provider/model selector
hermes model openrouter:anthropic/claude-opus-4 # Or do it in one line
Supported providers include:
- **OpenRouter** — 200+ models, one API key (best starting point)
- **Nous Portal** — Nous Research's own inference endpoint
- **OpenAI** — GPT-4o, o3, o4-mini
- **Anthropic** — Claude Opus 4, Sonnet 4
- **xAI** — Grok
- **Google Gemini** — Gemini 2.5 Pro
- **Hugging Face** — open models
- **NovitaAI**, **NVIDIA NIM**, **MiniMax**, **Kimi/Moonshot**, **z.ai/GLM**, **Xiaomi MiMo**
- **Local endpoints** — Ollama, vLLM, LM Studio, or any OpenAI-compatible API
Switch models mid-conversation with `/model`. Use different models for different tasks. Set up fallback chains so if one provider goes down, another takes over automatically. **No lock-in, no vendor risk.**
3. The Skills System: It Learns and Improves
This is Hermes' secret weapon. Skills are **on-demand knowledge documents** that the agent loads when needed — like procedural memory for an AI.
**How it works:**
- Hermes has a library of skills in `~/.hermes/skills/`
- Each skill is a markdown file with instructions for a specific task
- Skills use **progressive disclosure** — the agent only loads skill content when it needs it, saving tokens
- Built-in skills cover software development, research, creative work, system admin, and more
- **Hermes can create new skills autonomously** after completing complex tasks
- **Skills self-improve** — Hermes refines them during use
Real-world examples of built-in skills:
| Skill | What it does |
|---|---|
| `plan` | Creates implementation plans before coding |
| `github-pr-workflow` | Manages the entire PR lifecycle |
| `research-paper-writing` | Writes academic papers with templates |
| `p5js` | Creates generative art |
| `manim-video` | Produces animated explainer videos |
| `gif-search` | Searches and sends GIFs |
| `axolotl` | Fine-tunes LLMs with Axolotl |
| `blog-watcher` | Monitors RSS feeds and summarizes |
Skills are compatible with the [agentskills.io](https://agentskills.io) open standard, meaning the community can share, modify, and port skills across compatible agent frameworks.
4. Persistent Memory That Actually Works
Most chatbots forget everything the moment you close the tab. Hermes has a **bounded, curated memory system** that persists across sessions:
- **MEMORY.md** — General knowledge about your preferences, environment, and recurring patterns
- **USER.md** — A deepening model of who you are: your projects, your style, your goals
- **Session search** — FTS5 full-text search across all past conversations, with LLM summarization
- **Memory nudges** — Hermes periodically prompts itself to persist important information
- **External providers** — Plug in Honcho, Mem0, Supermemory, or other backends for advanced memory
After a few sessions, Hermes knows your project structure, your preferred tools, your coding style, and your communication preferences — without you having to re-explain everything every time.
5. Multi-Platform Gateway: Talk to It From Anywhere
Hermes isn't tied to your terminal. A single gateway process connects it to **20+ platforms** simultaneously:
- **Telegram** — Chat with Hermes from your phone
- **Discord** — AI agent in your server
- **Slack** — Workspace assistant
- **WhatsApp** — Personal AI on your phone
- **Signal** — Encrypted messaging
- **Matrix** — Decentralized chat
- **Mattermost** — Self-hosted team chat
- **Email / SMS** — Old school, still supported
- **DingTalk, Feishu, WeCom, Weixin, QQ Bot, Yuanbao, BlueBubbles**
- **Home Assistant** — Control your smart home
- **Microsoft Teams, Google Chat**
Start a task from Telegram, check progress on Discord, get results delivered to Slack. The gateway handles **cross-platform continuity** — Hermes knows who you are regardless of which app you're using.
6. Scheduled Automation (Cron)
Tell Hermes to do something on a schedule, and it will — even when you're not online:
# In chat:
/cron add "every 6h" "Check server health and alert if anything's wrong"
/cron add "daily at 9am" "Summarize new items from my RSS feeds" --skill blogwatcher
# From CLI:
hermes cron create "every 1h" "Back up project files" --skill backup
Cron jobs can:
- Run one-shot or recurring
- Attach zero, one, or multiple skills
- Deliver results to any configured platform
- Run in **no-agent mode** (pure script execution, zero LLM cost)
7. Subagent Delegation: Parallel Work
Need to research three topics simultaneously? Hermes can **spawn isolated child agents** that work in parallel:
delegate_task(tasks=[
{"goal": "Research Python async patterns", "toolsets": ["web"]},
{"goal": "Find the best CI/CD tools for 2026", "toolsets": ["web"]},
{"goal": "Draft a database migration plan", "toolsets": ["terminal", "file"]}
])
Each subagent gets a fresh context, restricted tools, and its own terminal. Only the summary enters the parent's context — saving tokens while maximizing throughput. Up to 3 concurrent subagents by default (configurable with no hard ceiling).
8. MCP Server Support
Hermes connects to **Model Context Protocol (MCP) servers** — the emerging standard for AI tool integration. This means you can plug in:
- **GitHub MCP** — Manage repos, issues, PRs
- **Database MCP** — Query Postgres, SQLite, MySQL
- **Filesystem MCP** — More granular file access
- **Any custom MCP server** you build
Per-server tool filtering lets you control exactly what Hermes can access.
Use Cases: What People Actually Build With Hermes
🖥️ Software Development
The most common use case. Hermes becomes your **AI pair programmer in the terminal**:
- **Code generation** — "Create a FastAPI service with auth, database models, and tests"
- **Debugging** — "The CI is failing on this test, figure out why" → runs it, reads logs, fixes it
- **Code review** — "Review this PR branch for security issues and performance problems"
- **Refactoring** — "Rename all instances of `old_api` to `new_api` across the codebase"
- **DevOps** — "Deploy the staging environment and run the smoke tests"
🔬 Research
- "Research the latest papers on multi-agent RL and give me a summary with citations"
- "Compare the architecture of llama.cpp vs vLLM"
- "Track this open-source project's issues and PR activity over the past week"
🛠️ System Administration
- "Check disk usage across all servers, find anything over 80%, and suggest cleanup"
- "Rotate my SSL certificates"
- "Monitor the logs for error rates and alert me on Telegram if they spike"
✍️ Content Creation
- With the `manim-video` skill: "Create an animated explainer video explaining how transformers work"
- With the `p5js` skill: "Generate a procedural art piece with flowing color gradients"
- With the `research-paper-writing` skill: "Draft an ACL-style paper on our findings"
🏠 Home Automation
Through the Home Assistant integration:
- "Turn off all lights when nobody's home"
- "What's the energy usage this month compared to last?"
- "If the temperature drops below 60°F, turn on the heating"
📈 Trading Bots
- "Run this trading strategy on a schedule and notify me on Telegram if it triggers"
- "Analyze these market conditions and write a report"
Real Examples From the Community
**"I have Hermes running on a $5 DigitalOcean droplet. It monitors my production servers, auto-deploys via GitHub Actions when tests pass, and sends me daily summaries on Telegram. I haven't SSH'd into that box in months."** — Senior DevOps Engineer
**"I built a full-stack Next.js app in an afternoon. Hermes scaffolded it, set up the database, wrote the API routes, and deployed it. I just reviewed the code and merged."** — Indie Hacker
**"My Hermes instance manages a Kanban board across my team. It auto-assigns tasks, checks progress daily, and escalates blocked items. It's basically a free project manager that never sleeps."** — Startup Founder
**"I use Hermes as a personal research assistant. It scours ArXiv, reads papers, cross-references citations, and writes literature reviews. What used to take me two weeks now takes two hours."** — PhD Researcher
Why Open Source Matters
Hermes is **MIT-licensed open source**. This isn't just a checkbox — it has real implications:
1. **Your data stays yours** — No data ever leaves your infrastructure unless you explicitly connect a third-party API. Everything runs locally.
2. **No rate limits, no tiers** — There's no "Pro" plan that unlocks tool calling or memory. Every feature is available to every user.
3. **You can inspect every line** — The entire codebase is on GitHub. You can audit security, understand behavior, and contribute fixes.
4. **No vendor lock-in** — If Nous Research disappears tomorrow, your Hermes keeps running. It's your software.
5. **Community-driven improvement** — 17,000+ tests, hundreds of contributors, active Discord community. The pace of development is staggering — new features ship weekly.
6. **The skills ecosystem** — Anyone can create, share, and install skills. The [Skills Hub](https://agentskills.io) is a growing marketplace of community-contributed capabilities.
The Hermes Ecosystem
| Resource | URL | Purpose |
|---|---|---|
| **GitHub** | [github.com/NousResearch/hermes-agent](https://github.com/NousResearch/hermes-agent) | Source code, issues, releases |
| **Documentation** | [hermes-agent.nousresearch.com/docs](https://hermes-agent.nousresearch.com/docs) | Full docs, guides, references |
| **Discord** | [discord.gg/NousResearch](https://discord.gg/NousResearch) | Community, support, show-and-tell |
| **Skills Hub** | [agentskills.io](https://agentskills.io) | Browse and share community skills |
| **LLMs.txt** | [docs/llms.txt](https://hermes-agent.nousresearch.com/docs/llms.txt) | Machine-readable doc index for AI coding tools |
The community on Discord is exceptionally active — developers share setups, troubleshoot issues, contribute skills, and showcase what they've built. It's the best place to see what Hermes can really do.
Key Takeaways
- **Hermes Agent is an autonomous AI agent** that lives in your terminal, not a web browser. It can execute commands, write files, search the web, and interact with APIs — not just answer questions.
- **It's provider-agnostic** — works with 20+ LLM providers including OpenRouter, OpenAI, Anthropic, and local models. No lock-in.
- **The skills system is a closed learning loop** — Hermes creates, improves, and reuses skills autonomously, getting smarter the more you use it.
- **Persistent memory** means it remembers you across sessions — your preferences, projects, and past conversations.
- **Multi-platform gateway** lets you talk to Hermes from Telegram, Discord, Slack, WhatsApp, Signal, and 15+ other platforms — all from a single process.
- **Cron jobs, subagent delegation, and MCP support** make it a true automation platform, not just a chat interface.
- **It's open source (MIT)** — self-hosted, no data leaves your machine, no vendor lock-in, no paid tiers.
- **A vibrant ecosystem** surrounds it: GitHub, Discord community, Skills Hub, and comprehensive documentation.
Quiz Questions
Test your understanding:
1. **What makes Hermes Agent fundamentally different from ChatGPT or Claude?**
- a) It has a better web interface
- b) It's a terminal-native agent with tool calling, persistent memory, and autonomous execution
- c) It only works with Nous Research models
- d) It requires a GPU to run
2. **What does "provider-agnostic" mean in the context of Hermes Agent?**
- a) It only uses one provider
- b) It can switch between 20+ LLM providers with no code changes
- c) It doesn't need any API keys
- d) It only works with open-source models
3. **How does the skills system keep Hermes from getting too expensive in token usage?**
- a) It caps the number of skills at 5
- b) It uses progressive disclosure — only loads skill content when needed
- c) It requires a subscription
- d) Skills are always loaded at startup
4. **Which of these is NOT a supported integration for the multi-platform gateway?**
- a) Telegram
- b) Discord
- c) Snapchat
- d) Home Assistant
5. **What license is Hermes Agent released under?**
- a) GPLv3
- b) Apache 2.0
- c) MIT
- d) Proprietary
**Answers:** 1-b, 2-b, 3-b, 4-c, 5-c
*Next Module: [Module 2: Installation & First Run](./module-02-installation.md)*